playbook-provision-prerelease.yml

- name: (linux) (prerelease) Install CrowdStrike Falcon and NRIA from prod bucket
  hosts: linux
  become: true
  gather_facts: no
  strategy: linear
  vars:
    ansible_aws_ssm_timeout: 900
    ansible_aws_ssm_retries: 10
  tasks:
    - name: Establish initial connection to host
      block:
        - name: Wait for connection to be available
          wait_for_connection:
            delay: 10
            timeout: 300
            connect_timeout: 60
          register: initial_connection

        - name: Verify connection with ping
          ansible.builtin.ping:
      rescue:
        - name: Reset connection on initial failure
          ansible.builtin.meta: reset_connection

        - name: Retry connection with extended timeout
          wait_for_connection:
            delay: 20
            timeout: 600
            connect_timeout: 120
          retries: 5
          delay: 60
          register: retry_connection
          until: retry_connection is succeeded

    - name: Gather facts for first time
      ansible.builtin.setup:
    
    - name: Ensure 'which' command is installed
      ansible.builtin.package:
        name: which
        state: present
      when: ansible_distribution == 'Rocky'

#    - name: Install crowdstrike falcon
#      ansible.builtin.include_role:
#        name: caos.ansible_roles.install_crowdstrike_falcon
#      vars:
#        falcon_client_id: "{{ lookup('env', 'CROWDSTRIKE_CLIENT_ID') }}"
#        falcon_client_secret: "{{ lookup('env', 'CROWDSTRIKE_CLIENT_SECRET') }}"
#        falcon_customer_id: "{{ lookup('env', 'CROWDSTRIKE_CUSTOMER_ID') }}"
#        api_base_url: "https://api.laggar.gcw.crowdstrike.com"

    - name: Install newrelic-infra agent with retry logic
      block:
        - name: Install newrelic-infra agent
          ansible.builtin.include_role:
            name: caos.ansible_roles.newrelic_cli
          vars:
            repo_endpoint: "https://nr-downloads-main.s3.amazonaws.com/"
            recipe: "newrelic-infra"
            region: "{{ lookup('ansible.builtin.env', 'NEW_RELIC_REGION') }}" # US | EU | STAGING
            nr_api_key: "{{ lookup('ansible.builtin.env', 'NEW_RELIC_API_KEY') }}"
            nr_account_id: "{{ lookup('ansible.builtin.env', 'NEW_RELIC_ACCOUNT_ID') }}"
      rescue:
        - name: Log installation failure for observability
          ansible.builtin.debug:
            msg: "WARNING: newrelic-infra installation failed on {{ inventory_hostname }}. Retrying in 30s..."

        - name: Wait before retry
          ansible.builtin.pause:
            seconds: 30

        - name: Retry newrelic-infra agent installation
          ansible.builtin.include_role:
            name: caos.ansible_roles.newrelic_cli
          vars:
            repo_endpoint: "https://nr-downloads-main.s3.amazonaws.com/"
            recipe: "newrelic-infra"
            region: "{{ lookup('ansible.builtin.env', 'NEW_RELIC_REGION') }}" # US | EU | STAGING
            nr_api_key: "{{ lookup('ansible.builtin.env', 'NEW_RELIC_API_KEY') }}"
            nr_account_id: "{{ lookup('ansible.builtin.env', 'NEW_RELIC_ACCOUNT_ID') }}"

        - name: Log successful retry
          ansible.builtin.debug:
            msg: "SUCCESS: newrelic-infra installation completed after retry on {{ inventory_hostname }}"

- name: Install fluent-bit from GH prerelease
  hosts: linux
  strategy: free
  vars:
    ansible_aws_ssm_timeout: 900
    ansible_aws_ssm_retries: 10
    node_version: 16.14.0
    # The following information is populated using the gathered inventory variables
    fluent_bit_package_name: "{{ tags.fb_package_name }}"
    pre_release_name: "{{ tags.pre_release_name }}"
    fb_version: "{{ tags.fb_version }}"
    os_distro: "{{ tags.os_distro }}"
    os_version: "{{ tags.os_version }}"
    arch: "{{ tags.arch }}"
    # Value set in versions/common.yml (nrFbOutputPluginVersion) and can be overridden via extra vars (-e nr_fb_output_plugin_version=<version>)
    nr_fb_output_plugin_version: "{{ nr_fb_output_plugin_version }}"
    # The following vars are injected in the fargate task env
    new_relic_api_key: "{{ lookup('ansible.builtin.env', 'NEW_RELIC_API_KEY') }}"
    new_relic_account_id: "{{ lookup('ansible.builtin.env', 'NEW_RELIC_ACCOUNT_ID') }}"
    new_relic_region: "{{ lookup('ansible.builtin.env', 'NEW_RELIC_REGION') }}"
  environment:
    NEW_RELIC_API_KEY: "{{ new_relic_api_key }}"
    NEW_RELIC_ACCOUNT_ID: "{{ new_relic_account_id }}"
    NEW_RELIC_REGION: "{{ new_relic_region }}"
  tasks:
    - name: Establish connection to host
      block:
        - name: Wait for connection to be available
          wait_for_connection:
            delay: 15
            timeout: 600
            connect_timeout: 60
      rescue:
        - name: Reset connection on failure
          ansible.builtin.meta: reset_connection

        - name: Retry connection with extended timeout
          wait_for_connection:
            delay: 20
            timeout: 600
            connect_timeout: 120
          retries: 5
          delay: 60
          register: retry_connection
          until: retry_connection is succeeded

    - name: Install Fluent Bit package to be tested for this distro
      ansible.builtin.include_role:
        name: install_fluent_bit_from_gh_prerelease
      vars:
        fb_package_name: "{{ fluent_bit_package_name }}"
        gh_prerelease_tag: "{{ pre_release_name }}"
      when: pre_release_name is not regex('^local-.*')

    - name: Install NR Fluent Bit output
      ansible.builtin.include_role:
        name: install_nr_fluent_bit_output
      vars:
        plugin_arch: "{{ (arch == 'x86_64' or arch == 'amd64') | ternary('amd64','arm64') }}"
        plugin_version: "{{ nr_fb_output_plugin_version }}"